Thanks for open-sourcing such as a wonderful work, could you please kindly help answer these questions?
The function “hypmlr” in hesp/utils/layers.py seems like contradictory to Eq. 7 in your paper Hyperbolic Image Segmentation. It looks like you did not divide ||wy|| when computing the input to sinh-1, also you did not multiply \lambda^{c}{p_y}.
Eq. 12 in your paper said that you multiply conditional probabilities along the ancestor to current descendant, however in function “get_joints” in hesp/embedding_space/abstract_embedding_space.py, It seems like you add them instead of multiplying, then in “decide” function you choose the maximum added conditionals as the prediction.
Thanks for open-sourcing such as a wonderful work, could you please kindly help answer these questions?
The function “hypmlr” in hesp/utils/layers.py seems like contradictory to Eq. 7 in your paper Hyperbolic Image Segmentation. It looks like you did not divide ||wy|| when computing the input to sinh-1, also you did not multiply \lambda^{c}{p_y}.
Eq. 12 in your paper said that you multiply conditional probabilities along the ancestor to current descendant, however in function “get_joints” in hesp/embedding_space/abstract_embedding_space.py, It seems like you add them instead of multiplying, then in “decide” function you choose the maximum added conditionals as the prediction.